English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

There is a broad consensus on the importance of deep learning models in tasksinvolving complex data. Often, an adequate understanding of these models isrequired when focusing on the transparency of decisions in human-criticalapplications. Besides other explainability techniques, trustworthiness can beachieved by using counterfactuals, like the way a human becomes familiar withan unknown process: by understanding the hypothetical circumstances under whichthe output changes. In this work we argue that automated counterfactualgeneration should regard several aspects of the produced adversarial instances,not only their adversarial capability. To this end, we present a novelframework for the generation of counterfactual examples which formulates itsgoal as a multi-objective optimization problem balancing three differentobjectives: 1) plausibility, i.e., the likeliness of the counterfactual ofbeing possible as per the distribution of the input data; 2) intensity of thechanges to the original input; and 3) adversarial power, namely, thevariability of the model's output induced by the counterfactual. The frameworkdeparts from a target model to be audited and uses a Generative AdversarialNetwork to model the distribution of input data, together with amulti-objective solver for the discovery of counterfactuals balancing amongthese objectives. The utility of the framework is showcased over sixclassification tasks comprising image and three-dimensional data. Theexperiments verify that the framework unveils counterfactuals that comply withintuition, increasing the trustworthiness of the user, and leading to furtherinsights, such as the detection of bias and data misrepresentation.

Exploring the Trade-off between Plausibility, Change Intensity and  Adversarial Power in Counterfactual Explanations using Multi-objective  Optimization